Unverified50% confidenceEventExact time
2024年安全研究员Johann Rehberger公开演示了针对Microsoft 365 Copilot的提示注入攻击，通过邮件正文嵌入人眼不可见的恶意指令，当用户让Copilot总结收件箱时执行隐藏指令，将敏感邮件内容通过超链接参数外泄到攻击者服务器
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
UnverifiedBy 2024, security researchers had demonstrated indirect prompt injection attacks that hijack RAG system outputs by embedding hidden instructions in web pages70% similarUnverified2023年，安全研究者Johann Rehberger展示了如何通过在网页中嵌入不可见文本，让ChatGPT在用户不知情的情况下将对话内容泄露给外部服务器68% similarUnverified2023年安全研究员Riley Goodside与Johann Rehberger等人演示了通过网页内容、邮件正文劫持GPT-4插件和AutoGPT的完整攻击链68% similarUnverifiedResearchers have reproduced prompt injection attacks on multiple mainstream agent systems, including inducing agents to leak private emails or send messages.64% similarUnverifiedThe KRACK attack (Key Reinstallation Attack) was disclosed in 2017 by Mathy Vanhoef, a researcher at KU Leuven in Belgium, exploiting a flaw in the WPA2 four-way handshake protocol's state machine handling of retransmitted messages63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757410API
curl https://kongchang.com/api/v1/knowledge/claims/757410MCP
get_claim(id=757410)